Reconhecimento facial com super-resolução: uma abordagem utilizando redes generativas e joint-learn

Abstract

Surveillance cameras are broadly used in supervising private places to restrain violent acts. One of the ways of improving this system is recognizing people in this space, preferably by using an individual’s face biometrics. An existing challenge is to recognize faces when imaging conditions are adverse, either by low-quality cameras or the distance between the subject and the camera, thus impacting the accuracy of these recognizing systems. Super-Resolution (SR) techniques can be used to improve both image resolution and quality before recognizing the face, to improve the accuracy of the recognition task. Among these techniques, the actual State of the Art uses Generative Adversarial Networks (GAN). When used together, one promising option is to train Super-Resolution and Face Recognition as one single network, conducting the network to learn SR features that will improve its capability when recognizing faces. In the present work, we trained a Super-Resolution Face Recognition model using this joint-learn approach, combining a Generative network for SR, and a ResNet50 for Face Recognition. The model was trained with a Discriminator network, following the GAN training framework. The images generated by the network were convincing, but we couldn’t converge the FR model in a timely manner. We hope that our contributions could help future works on this topic. Code is publicly available at https://github.com/OliRafa/SRFR-GAN.